Real-time non-rigid target tracking for ultrasound-guided clinical interventions

Abstract

Biological motion is a problem for non- or mini-invasive interventions when conducted in mobile/deformable organs due to the targeted pathology moving/deforming with the organ. This may lead to high miss rates and/or incomplete treatment of the pathology. Therefore, real-time tracking of the target anatomy during the intervention would be beneficial for such applications. Since the aforementioned interventions are often conducted under B-mode ultrasound (US) guidance, target tracking can be achieved via image registration, by comparing the acquired US images to a separate image established as positional reference. However, such US images are intrinsically altered by speckle noise, introducing incoherent gray-level intensity variations. This may prove problematic for existing intensity-based registration methods. In the current study we address US-based target tracking by employing the recently proposed EVolution registration algorithm. The method is, by construction, robust to transient gray-level intensities. Instead of directly matching image intensities, EVolution aligns similar contrast patterns in the images. Moreover, the displacement is computed by evaluating a matching criterion for image sub-regions rather than on a point-by-point basis, which typically provides more robust motion estimates. However, unlike similar previously published approaches, which assume rigid displacements in the image sub-regions, the EVolution algorithm integrates the matching criterion in a global functional, allowing the estimation of an elastic dense deformation. The approach was validated for soft tissue tracking under free-breathing conditions on the abdomen of seven healthy volunteers. Contact echography was performed on all volunteers, while three of the volunteers also underwent standoff echography. Each of the two modalities is predominantly specific to a particular type of non- or mini-invasive clinical intervention. The method demonstrated on average an accuracy of ∼1.5 mm and submillimeter precision. This, together with a computational performance of 20 images per second make the proposed method an attractive solution for real-time target tracking during US-guided clinical interventions.

Anatomically plausible models and quality assurance criteria for online mono- and multi-modal medical image registration

“…Therefore the results indicate that, in practice, as long as the acquisition sequence parameters do not change considerably, the algorithm parameters do not require re-calibration. This observation is also in good correspondence with previous studies (Zachiu et al 2015, Zachiu et al 2015b, Denis de Senneville et al 2016, Zachiu et al 2017a, Zachiu et al 2017b).…”

“…However, the transition towards the contrast-devoid area is resolved with high divergences in the estimated deformations, leading to strong misregistrations in the ventral part of the liver. Nevertheless, if sufficient contrast is available within the images, the EVolution algorithm was demonstrated to be capable of high accuracy and precision in several independent studies (Denis de Senneville et al 2016, Zachiu et al 2017a, Zachiu et al 2017b. Concerning the required computational time, the EVO algorithm introduces a rather large relative penalty, compared to the HSO and HSI.…”

Accelerating multi-modal image registration using a supervoxel-based variational framework

“…We also recall that it is common to accelerate the registration process by simply running the algorithm on down-sampled versions of the two input images (Glitzner et al 2015, Zachiu et al 2017c. The obtained deformation field is then upsampled at the original image resolution and the magnitude of each displacement vector is adjusted accordingly.…”

“…In addition, EVolution also offers specific tracking capability in scenarios for which structures in the image to register do not have counterparts in the reference image. The EVolution algorithm was recently tested with various application scenarios such as T1/ T2-MR (Denis de Senneville et al 2016) and CT/CBCT registration (Zachiu et al 2017b), as well as real-time target US-based tracking dedicated to HIFU therapies (Zachiu et al 2017c).…”